Chinese Internet AS-Level Topology - arXivactually travels toward a destination network and they rarely include Internet exchange points (IXP) ASes. Traceroute measurements also have

Chinese Internet AS-Level Topology

Shi Zhou

Department of Computer Science University College London

Ross Building Adastral Park Ipswich IP5 3RE United Kingdom

email szhouadastraluclacuk

Guo-Qiang Zhang and Guo-Qing Zhang

Institute of Computing Technology Chinese Academy of Sciences

Ke-xue-yuan South Road Beijing 100080 China

email guoqiangictaccn gqzhangictaccn

February 1 2008

Abstract

We present the first complete measurement of the Chinese Internet topology at

the autonomous systems (AS) level based on traceroute data probed from servers of

major ISPs in mainland China We show that both the Chinese Internet AS graph and

the global Internet AS graph can be accurately reproduced by the Positive-Feedback

Preference (PFP) model with the same parameters This result suggests that the

Chinese Internet preserves well the topological characteristics of the global Internet

This is the first demonstration of the Internetrsquos topological fractality or self-similarity

performed at the level of topology evolution modeling

1 Introduction

During the last three decades the Internet has experienced fascinating evolution both

exponential growth in traffic and endless expansion in topology It is crucial to obtain a

good description of the topology because effective engineering of the Internet is predicated

on a detailed understanding of issues such as the large-scale structure of its underlying

physical topology the manner in which it evolves over time and the way in which its

constituent components contribute to its overall function [1]

The Internet contains millions of routers which are grouped into thousands of sub-

networks called autonomous systems (AS) Network structure inside an AS only affect

local properties whereas the delivery of data traffic through the global Internet depends

on the complex interactions between ASes that exchange routing information using the

Border Gateway Protocol (BGP) [2] This paper focuses on the Internet AS-level topol-

ogy on which a node is an AS and a link represents a BGP peering relation between two

ASes Measurements of the Internet AS-level topology [3 4 5] became available in late

1990rsquos These measurements have greatly improved our understanding of the evolution

and structure of the Internet [6]

During the last decade the Internet has developed rapidly in China However mea-

surement of the Chinese Internet topology has only been carried out from outside of the

country In this paper we present the first and so far the most complete measurement

of the Chinese Internet AS graph based on traceroute data probed from servers of major

ISPs inside mainland China The obtained Chinese Internet AS graph is a small regional

subgraph of the massive global Internet AS graph because it contains only ASes located

in mainland China We show that both the global graph and the small local graph can

be accurately reproduced by the Positive-Feedback Preference (PFP) model [7 8] with

the same parameters This result suggests that the Chinese Internet preserves well the

topological characteristics of the global Internet Our work demonstrates the existence of

a statistical fractality in the description of the Internet AS-level topology

2 Measurements of Internet Topology

There are two primary methods for inferring Internet AS-level topology one is the passive

measurements inferring AS connectivity information from BGP routing tables eg [3 4]

and the other is the active measurement which extracts AS graph from traceroute probing

data eg [5] AS graphs produced by the two methods exhibit a number of different topol-

ogy properties [9] The BGP measurements reflect the control plane of the AS topology

whereas the traceroute measurements reflect the data plane of the AS topology BGP

tables have the advantage that they are relatively easy to parse process and comprehend

however BGP measurements suffer certain limitations in that they do not reflect how traffic

actually travels toward a destination network and they rarely include Internet exchange

points (IXP) ASes Traceroute measurements also have inherent limitations associated

with converting IP addresses (on the router level) to AS numbers however traceroute

measurements reflects the actual paths that IP packets follow In this paper we focus on

the active measurements based on traceroute data

21 The Global Internet AS Graph ndash ITDK0304

In 1998 CAIDA [5] launched the Macroscopic Topology Project to collect and analyze

Internet-wide topology at a representatively large scale CAIDArsquos topology measurement

tool skitter implements the Internet Control Message Protocol (ICMP) It collects the

forward path from the monitor to a given destination and captures the addresses of in-

termediate routers in the path The skitter runs on more than 20 monitors around the

globe and actively collects forward IP path to over half a million destinations across the

current IPv4 address space CAIDA extracts interconnect information of ASes from the

massive traceroute data collected by skitter and produces AS adjacency graphs CAIDA

argues [10] that while inherent limitations of the traceroute-based probing methodology

do not allow for 100-accurate extraction of the real Internet topology from skitter data

CAIDA seeks data sources that are collectively most likely to capture a precise and coher-

ent snapshot of macroscopic Internet structure We extract the global Internet AS graph

in this paper from the Internet topology data kit ITDK0304 collected by CAIDA in April

2003 [11] The ITDK0304 Internet AS graph contains 9024 nodes and 28959 links

22 The Chinese Internet AS Graph ndash CN05

The Internet in China has developed rapidly in the last decade Today the country has

more than 100 million Internet users [12] However measurement of the Chinese Internet

topology has only been carried out from outside of the country For example currently

there is no skitter monitor located in China

Recently we measured the Chinese AS graph using the active measurement method

We collected traceroute data from six servers belonging to the following major ISPs in

mainland China CSTNET (China Science and Technology Net) CERNET (China Ed-

ucation and Research Net) CAPINFO Company Limited ChinaTelecom ChinaNetCom

and CGWNET (China Great Wall Net) The measurement process took place during the

first week in May 2005

From each server a traceroute tool probes 7436 destination IP addresses in mainland

China including (i) 3010 Web server addresses which are obtained from the Chinese Web

yellow pages database [13] and (ii) 4426 random IP addresses which are obtained by

using the stub network sampling method [14] which samples one IP address from every

IP prefix assigned to mainland China by APNIC [15] The traceroute tool probes each IP

address for three times Although all the Web server addresses are reachable by HTTP

they are not 100 reachable by the traceroute tool because many Web servers and ASes

do not allow ICMP probing from external servers Of the 3010 Web server addresses

the traceroute tool reached 1859 addresses and ASes to which 2753 addresses belong Of

the 4426 random IP addresses the traceroute tool reached 336 addresses and ASes of

3485 addresses The success rate of probing random IP addresses is even lower because

many randomly sampled IP addresses do not exist We observed that the size of obtained

Chinese AS graph will not increase significantly when probing more destination addresses

We accumulated all the traceroute data and used the same techniques of CAIDA [10]

to convert IP addresses to corresponding AS numbers announced by APNIC The resulted

Chinese AS graph is called CN05 which contains 84 nodes and 211 links The dataset is

available at httpwwwadastraluclacuksimszhouresourcehtm in the format of

a list of peering AS numbers Our measurement is more complete than the Chinese AS

graph extracted from CAIDArsquos ITDK0304 which contains only 71 Chinese ASes and 160

peering links among them

3 Comparison Between The Two AS Graphs

The Chinese Internet AS graph CN05 is a small subgraph of the global Internet AS graph

ITDK0304 In this paper we compare the two AS graphs in an indirect way ie we contrast

the two AS graphs against networks generated by the Positive-Feedback Preference (PFP)

model [7 8] which has been regarded by CAIDA as the most precise and complete Internet

topology generator to date [16 17 18]

31 The Positive-Feedback Preference Model

The PFP model is an extensive modification of the Barabasi-Albert model [19] The PFP

model grows Internet-like networks by using the following two generative mechanisms

namely the interactive growth and the positive-feedback preference which are coincident

with a number of observations on the Internet history data [20 21 22]

Interactive Growth Starting from a small random graph at each time step 1) with

probability p isin [0 1] (see Fig 1a) a new node is attached to an old node in the existing

system and then two new links are added connecting the old node to two other old nodes

and 2) with probability 1 minus p (see Fig 1b) a new node is attached to two old nodes and

then one new link is added connecting one of the two old nodes to another old node

Positive-Feedback Preference Degree k is defined as the number of links a node has

The preference probability that a new link is attached to node i is given as

Π(i) =k

1+δ log10 ki

j k1+δ log10 kj

δ isin [0 1] (1)

in which a nodersquos ability of acquiring new links increases as a feed-back loop of the node

degree Such a preference scheme results in ldquothe rich not only get richer but get dispro-

portionately richerrdquo

Numerical simulation shows that the PFP model produces the best result when the

parameters p = 04 and δ = 0048 [7 8] As stated in the status report of IRTF Routing

Research Grouprsquos (RRG) Future Domain Routing (FDR) Scalability Research Subgroup

Table 1 AS Graphs vs PFP Networks

CN05 PFP1 ITDK0304 PFP2

Number of nodes N 84 84 9204 9204

Number of links L 211 217 28959 27612

Maximum degree kmax 38 39 2070 1950

Degree exponent γ -221 -221 -2254 -2255

Assortative coefficient α -0328 -0298 -0236 -0234

Top clique size nclique 3 3 16 16

Rich-club exponent θ -142 -142 -148 -149

Average triangle coefficient 56 68 214 191

Characteristic path length ℓlowast 254 254 312 307

Maximum coreness cmax 5 6 28 27

(RR-FS) in 2005 [18] the PFP model ldquois the best non-equilibrium network growth model

with respect to its proximity to observed Internet topologyrdquo

32 Simulation Result

We use the PFP model to grow networks to the same number of nodes as CN05 and

ITDK0304 (see Table 1) using the same model parameters as suggested in previous nu-

merical simulations [7 8] ie p = 04 and δ = 0048 For each AS graph ten networks

are generated using random seeds and their properties are averaged

The Internet topology is extremely complex Researchers have proposed a number of

metrics to quantify and explain the structure of Internet [6] In this paper we compare

the AS graphs and the PFP networks by examining the following topological properties

1 Degree distribution

2 Degree correlation

3 Rich-club connectivity

4 Triangle coefficient

5 Shortest path length

6 k-core structure

Note that CN05 is a very small graph and therefore all its statistics are of very low

statistical significance

321 Degree Distribution

Degree is a local property but the probability distribution of degree provides a view on

the global structure of a network Faloutsos et al [23] showed that the Internet AS-level

topology exhibits a power-law degree distribution

P (k) sim kγ γ ≃ minus22

This means a few nodes have very large numbers of links however the vast majority of

nodes have only a few links This discovery is significant because it invalidated previous

Internet models that were based on the classical random graphs having Poisson degree

distributions

Fig 2 shows that ITDK0304 and CN05 are characterized by a power-law degree distri-

bution (Degree distribution of CN05 is not statistically significant due to its small number

of nodes) with the power-law exponent of CN05 γ = minus221 is slightly larger than that of

the ITDK0304 γ = minus225 (see Table 1) It is clear that the PFP networks closely match

the degree properties of the two AS graphs

As [24] pointed out the degree distribution alone does not uniquely characterize a

network topology To describe the full picture of network structure we need to examine

other important topological properties as well

322 Degree Correlation

Recently a number of studies have shown that the degree correlation or the joint degree

distribution P (k kprime) defined as the probability that a link connects k- and kprime-degree

nodes plays a significant role in defining a networkrsquos structure [9 25 26 27] Networks

exhibit different mixing patterns according to their degree correlation Social networks are

assortative mixing where nodes tend to attach to alike nodes ie high-degree nodes to high-

degree nodes and low-degree nodes to low-degree nodes On contrast technological and

biological networks including the Internet are dissasortative mixing where high-degree

nodes tend to connect with low-degree nodes and visa versa

A networkrsquos mixing pattern can be indicated by the correlation between a nodersquos degree

k and its nearest-neighbors average degree knn [22] A mixing pattern can also be identified

by measuring the assortative coefficient [25] minus1 lt α lt 1 which is defined as

α =Lminus1 sum

m jmkm minus [Lminus1 summ

12(jm + km)]2

Lminus1sum

m12(j2

m + k2m) minus [Lminus1

12(jm + km)]2

where L is the number of links a network has and jm km are the degrees of the nodes at

the ends of the mth link with m = 1 L If 0 lt α lt 1 a network is assortative mixing

and if minus1 lt α lt 0 a network is dissasortative mixing

Fig 3 shows that ITDK0304 and CN05 clearly exhibit a negative correlation between

degree and the nearest-neighbors average degree Also as shown in Table 1 both the AS

graphs are characterized by a negative assortative coefficient The PFP networks precisely

reproduce the disassortative mixing properties of the Internet AS graphs

323 Rich-Club Connectivity

A hierarchy property of the Internetrsquos structure is the rich-club phenomenon [28] which

describes the fact that well connected nodes rich nodes are tightly interconnected with

other rich nodes forming a core group or rich-club The rich-club phenomenon does not

imply that the majority links of the rich nodes are directed to other club members Indeed

rich nodes have very large numbers of links and only a few of them are enough to provide

the interconnectivity to other club members whose number is anyway small

The rich-club membership can be defined as ldquo nodes with degree no smaller than k rdquo or

ldquo the r best connected nodes rdquo where rank r = 1N denotes a nodersquos position on the non-

increasing degree list of a graph of size N The rich-club phenomenon is quantitatively

assessed by the metric of rich-club connectivity ϕ which is defined as the ratio of the

actual number of links to the maximum possible number of links among members of a

rich-club Rich-club connectivity measures how well club members ldquo knowrdquo each other

eg ϕ = 1 means that all the members have a direct link to any other member ie

they form a fully connected mesh a clique The top clique size nclique is defined as the

maximum number of highest rank nodes still forming a clique

Fig 4 illustrates the rich-club connectivity as a function of node degree It shows that

a node in ITDK0304 should have a degree larger than 200 to be a member of the top

clique whereas a node in CN05 can be a member of the top clique with a degree of merely

20 Fig 5 illustrates the rich-club connectivity as a function of node rank normalized by

the network size N We can see that both CN05 and ITDK0304 obey a power law of

ϕ(rN) sim (rN)θ with the exponent θ of minus142 and minus148 respectively (see Table 1) The

PFP networks accurately resemble all these rich-club properties of the two AS graphs

324 Triangle Coefficient

Short cycles (eg triangles and quadrangles) encode the redundancy information in a

network structure because the multiplicity of paths between any two nodes increases with

the density of short cycles The triangle coefficient kt is defined as the number of triangles

that a node shares [7] Different from the widely-studied clustering coefficient [29] which

can be given as c = kt

k(kminus1)2 the triangle coefficient is able to infer neighbor clustering

information of nodes with different degrees

Fig 6 shows the complementary cumulative distribution (CCD) of triangle coefficient

The two AS graphs exhibit similar power-law distributions Fig 7 shows that the corre-

lation between degree and triangle coefficient of the two AS graphs are nearly overlapped

with each other ndash the larger degree a node has the more triangles it shares The PFP

networks closely match these triangular properties of the two AS graphs

325 Shortest Path Length

The Internet is a small-world network [29] because it is possible to get to any node via

only a few links among adjoining nodes The shortest path length ℓ is the minimum hop

distance between a pair of nodes The characteristic path length ℓlowast of a network is the

average of shortest path lengths over all pairs of nodes Performance of modern routing

algorithms depend strongly on the distribution of shortest path [30]

Fig 8 shows the CCD of shortest path length It is clear that CN05 is smaller than

ITDK0304 (also see ℓlowast in Table 1) Fig 9 shows that the two AS graphs show a negative

correlation between the shortest path length and node degree ie in general the higher

degree a node has the shorter average distance between the node to all other nodes Again

the PFP networks precisely resemble the shortest path properties of the two AS graphs

326 k-core Structure

Recently researchers are interested in a topological property called the k-core [31] As

shown in Fig 10 a k-core of a network can be obtained by recursively removing all the

nodes of degree less than k until all nodes in the remaining graph have at least degree k

A node has the coreness value of c if it belongs to the c-core but not to (c + 1)-core

Nodes having the same coreness value of c are called the c-shell of a network The k-

core decomposition by recursively pruning the least connected nodes disentangles the

hierarchical structure of a network by progressively focusing on its central cores

Fig 11 shows visualizations of the networks k-core structure Pictures are produced

by the tool LaNet-VI [32 33] On each picture a network node is illustrated as a dot

whose size is exponentially proportional to the nodersquos degree and whose color represents

the nodersquos coreness value Dots are distributed on a series of concentric circular c-shells

corresponding to their coreness value c The diameter of each shell depends on the shell

index c and is proportional to cmax minus c The pictures clearly show that the PFP networks

closely resemble the k-core structure of the two AS graphs

33 Discussion

The above analysis demonstrates that the Chinese Internet AS graph preserves well the

structural characteristics of the global Internet AS graph The Internet in China has

developed in a social-economic environment that is characterized by more centralized

planning and less commercial competition than the global Internet dominated by the

West However this does not make the evolution of the Internet AS-level topology in

China different from the West because the local social-economic factors mainly affect

the development of the last-mile access networks eg China has a small number of IPSs

but all ISPs are very large Whereas the evolution of the Internet AS-level topology is

primarily influenced by technological factors which are fairly universal That is to say

when making a decision on a BGP peering relationship [34] all ISPs around the world

tend to consider the same technical issues such as performance objectives and technical

constrains In our future work we will investigate the Internet AS-level subgraphs in other

countries and regions

4 Conclusion

In this paper we present the first traceroute measurement of the Chinese Internet AS-level

topology collected from probing servers inside the country We compare the global Internet

AS graph with the Chinese Internet AS graph using the PFP model as a yardstick We

examine a diversified set of topological characteristics of the two AS graphs including

degree distribution degree correlation rich-club connectivity triangle coefficient shortest

path length and k-core structure Remarkably the same parameters in the PFP model to

describe the global Internet AS graph also describe the Chinese Internet AS graph This is

the first demonstration of the Internetrsquos topological fractality or self-similarity performed

at the level of topology evolution modeling

5 Acknowledgments

This work is partly supported by the UK Nuffield Foundation under Grant No NAL01125G

and the National Natural Science Foundation of China under Grant No 60673168 The

Internet data kit ITDK0304 is provided by CAIDA

References

[1] Floyd S and Kohler E lsquoInternet research needs better modelsrsquo Comput Commun

Rev 2003 33 (1) pp 29ndash34

[2] Quoitin B Pelsser C and Swinnen L lsquoInterdomain traffic engineering with BGPrsquo

IEEE Communications Magazine 2003 41 (5) pp 122ndash128

[3] National Laboratory for Applied Network Research httpmoatnlanrnet ac-

cessed November 2006

[4] Route Views Project University of Oregon Eugene httpwwwrouteviewsorg

accessed November 2006

[5] The Cooperative Association For Internet Data Analysis (CAIDA) San

Diego Supercomputer Center University of California San Diego USA

httpwwwcaidaorg accessed November 2006

[6] Pastor-Satorras R and Vespignani A lsquoEvolution and Structure of the Internet -

A Statistical Physics Approachrsquo (Cambridge University Press 2004)

[7] Zhou S and Mondragon R J lsquoAccurately modelling the Internet topologyrsquo Phys

Rev E 2004 70 (066108)

[8] Zhou S lsquoUnderstanding the evolution dynamics of Internet topologyrsquo Phys Rev E

2006 74 (016124)

[9] Mahadevan P Krioukov D Fomenkov M Huffaker B Dimitropoulos X claffy

kc and Vahdat A lsquoThe Internet AS-level topology Three data sources and one

definitive metricrsquo Comput Commun Rev 2006 36 (1) pp 17ndash26

[10] Hyun Y Broido A and claffy kc lsquoTraceroute and BGP AS path incongruitiesrsquo

httpwwwcaidaorgoutreachpapers2003ASP accessed November 2006

[11] CAIDA Internet topology data kit ITDK0304

httpwwwcaidaorgtoolsmeasurementskitteridkdataxml accessed

November 2006

[12] China Internet Network Information Center lsquoThe 18th Sta-

tistical Report on the Internet Development in Chinarsquo 2006

httpwwwcnnicnetcnenindex0Oindexhtm accessed November 2006

[13] Chinese Web Yellow Pages httpwwwchinaypcomcn accessed November 2006

[14] Jinag Y and He S lsquoStudy on the destination selection method in router-

level Internet topology measurementrsquo Journal on Communications 2006 27 (2)

httppact518hiteducnsimjypdfsjc2702-p29pdf accessed November 2006

[15] Asia Pacific Network Information Center (APNIC) httpwwwapnicnet ac-

[16] IRTF Routing Research Grouprsquos (RRG) Future Domain Routing (FDR)Scalability

Research Subgroup (RR-FS) httprr-fscaidaorg accessed November 2006

[17] CAIDArsquos NeTS-NR Project lsquoToward Mathematically Rigourous Next-

Generation Routing Protocols for Realistic Network Topologiesrsquo 2004

httpwwwcaidaorgfundingnets-nr accessed November 2006

[18] Krioukov D and Krapivsky P lsquoPower Laws

as a Pre-asymptotic Regime of the PFP Modelrsquo

httpwwwcaidaorgpublicationspresentations2006isma0605 dimaisma0605 dimapdf

[19] Barabasi A L and Albert R lsquoEmergence of scaling in random networksrsquo Science

1999 286 pp 509ndash512

[20] Chen Q Chang H Govindan R Jamin S Shenker S J and Willinger W

lsquoThe origin of power laws in Internet topologies (revisited)rsquo Proc of INFOCOM

2002 pp 608ndash617

[21] Park S T Khrabrov A Pennock D M Lawrence S Giles C L and Ungar

L H lsquoStatic and dynamic analysis of the Internetrsquos susceptibility to faults and

attacksrsquo Proc of INFOCOM 2003 3 pp 2144ndash2154

[22] Vazquez A Pastor-Satorras R and Vespignani A lsquoLarge-scale topological and

dynamical properties of Internetrsquo Phys Rev E 2002 65 (066130)

[23] Faloutsos M Faloutsos P and Faloutsos C lsquoOn powerndashlaw relationships of the

Internet topologyrsquo Comput Commun Rev 1999 29 pp 251ndash262

[24] Tangmunarunkit H Govindan R Jamin S Shenker S and Willinger W lsquoNet-

work topology generators Degree-based vs structuralrsquo Proc of SIGCOMM 2002

pp 147ndash159

[25] Newman M E J lsquoMixing patterns in networksrsquo Phys Rev E 2003 67 (026126)

[26] Vazquez A M Boguna Y Moreno R P-S and Vespignani A lsquoTopology and

correlations in structured scale-free networksrsquo Phys Rev E 2003 67 (046111)

[27] Maslov S Sneppen K and Zaliznyak A lsquoDetection of topological patterns in

complex networks correlation profile of the Internetrsquo Physica A 2004 333 pp 529

[28] Zhou S and Mondragon R J lsquoThe rich-club phenomenon in the Internet topologyrsquo

IEEE Comm Lett 2004 8 (3) pp 180ndash182

[29] Watts D J and Strogatz S H lsquoCollective dynamics of lsquosmall-worldrsquo networksrsquo

Nature 1998 393 pp 440

[30] Labovitz C Ahuja A Wattenhofer R and Venkatachary S lsquoThe impact of

Internet policy and topology on delayed routing convergencersquo Proc of INFOCOM

2001 pp 537ndash546

[31] Dorogovtsev S N Goltsev A V and Mendes J F F lsquok-core organization of

complex networksrsquo Phys Rev Lett 2006 96 (040601)

[32] Alvarez-Hamelin J-I DallrsquoAsta L Barrat A and Vespignani A lsquok-

core decomposition a tool for the visualization of large scale networksrsquo

httparxivorgabscs0511007 accessed November 2006

[33] Topology visualisation tool LaNet-VI httpxavierinformaticsindianaedulanet-vi

[34] Chang H Jamin S and Willinger W lsquoTo Peer or not to Peer Modeling the

Evolution of the Internetrsquos AS-level Topologyrsquo Proc of INFOCOM 2006

Figure captions

1 Interactive growth of the PFP model

3 Nearest-neighbors average degree of k-degree nodes

4 Rich-club connectivity vs degree

5 Rich-club connectivity vs normalized rank

6 Complementary cumulative distribution (CCD) of triangle coefficient

7 Average triangle coefficient of k-degree nodes

8 Complementary cumulative distribution (CCD) of shortest path length

9 Correlation between shortest path length and degree

10 Example of k-core decomposition

11 Visualization of k-core structure

Figure 1 Interactive growth of the PFP model

Figure 2 Degree distribution

Figure 3 Nearest-neighbors average degree of k-degree nodes

Figure 4 Rich-club connectivity vs degree

Figure 5 Rich-club connectivity vs normalized rank

Figure 6 Complementary cumulative distribution (CCD) of triangle coefficient

Figure 7 Average triangle coefficient of k-degree nodes

Figure 8 Complementary cumulative distribution (CCD) of shortest path length

Figure 9 Correlation between shortest path length and degree

Figure 10 Example of k-core decomposition

(a) Internet AS graph (ITDK0304) (b) PFP model (N=9024)

(c) Chinese Internet AS graph (CN05) (d) PFP model (N=84)

Figure 11 Visualization of k-core structure

Introduction

Measurements of Internet Topology

The Global Internet AS Graph -- ITDK0304

The Chinese Internet AS Graph -- CN05

Comparison Between The Two AS Graphs

The Positive-Feedback Preference Model

Simulation Result

Degree Distribution

Degree Correlation

Rich-Club Connectivity

Triangle Coefficient

Shortest Path Length

k-core Structure

Discussion

Conclusion

Acknowledgments